Introducing Baseten Loops: A Training SDK for Frontier RL. Learn more here

Changelog

See our latest feature releases, product improvements and bug fixes

May 7, 2026

View configs for deployed models

truss model-config --model-id <id> --deployment-id <id> prints the YAML config of a deployed model, returning the original config.yaml when available. Add --output json for the full structured...

May 7, 2026

Browser-based login for Truss

truss login now provides the option to authenticate in your browser using your Baseten login. The truss auth group adds login, logout, and status subcommands; use --remote <name> to specify a remote...

May 5, 2026

Environment-scoped logs and metrics

View logs and metrics for every deployment in an environment from one place. Select an environment in the Logs or Metrics tab to scope telemetry to that environment instead of a single deployment.

Apr 17, 2026

Model API deprecation notice (DeepSeek v3 0324, GLM 4.6)

The DeepSeek v3 0324 and GLM 4.6 Model API(s) will be deprecated at 5pm PT on May 1st.

Apr 16, 2026

Cache token pricing now available for Model APIs

Cached input tokens are billed at a discounted rate on Model APIs for all models (excluding GPT-OSS), starting April 17, 2026. Cache token pricing is applied automatically to the portion of each...

Apr 6, 2026

Copy and download logs

You can now copy or download all visible logs directly from the logs viewer. A new export menu next to the search box lets you copy logs to your clipboard, or download them as CSV or JSON. To export...

Apr 6, 2026

Named entity recognition on BEI-Bert

BEI-Bert now supports token-classification models for named-entity recognition. Deploy any ForTokenClassification model with the /predict_tokens endpoint and get structured entity predictions with...

Apr 1, 2026

Per-request log filtering

Every predict call now returns a unique request ID in the X-Baseten-Request-Id response header. Use this ID to filter your model's logs to a single request, cutting through the noise when debugging...

Mar 31, 2026

Health check improvements

Startup probes now handle initialization more reliably by waiting until the model has loaded before executing any liveness checks. The startup phase still defaults to 30 minutes and can be configured...

Mar 30, 2026

Rolling deployments

You can now gradually shift traffic to new deployments instead of swapping all at once. Candidate replicas scale up incrementally while previous replicas scale down in controlled steps, giving you...